Future city, always rain. Man with sad eyes hunt machine people who look like real people. Machine people want more time to live, not happy being throw away. Big building man make machine people, but no heart for their wanting. Sad eyes man find machine lady, she not know she machine. They run, but other machine people want live more, find sad eyes man. Big fight on roof, rain pouring, machine man save sad eyes man, why? He just want see things, feel life. Bird fly in rain, machine man stop working. Sad eyes man think, what mean to be alive? Machine lady and he try escape, but future dark, uncertain. Is love real if heart beep by battery?

Big foot tiny man find shiny circle thing in dirt. Old man with pointy hat say circle thing is big trouble, must go far away to fire mountain for destroy. Tiny man and friends - some tiny, some tall, one shiny circle lover - go on big walk with trees and scary flying things. They meet beard men, arrow boy, and shiny sword man, all join for adventure. Big eye always watching, wanting shiny circle back. Lots of walking, fighting with green ugly creatures, and tiny man getting tired. Big spider scare, shiny circle lover betray, but fire mountain close. Can tiny man throw shiny circle in fire? Big fight, eagle save, and long boat trip for pointy hat man. Who will be hero? Shiny circle big problem, but friendship bigger.

I’m using LumaAI’s genie on their Discord, works well, really fast to get a preview and then takes around 20min for a refined version. Still early days of generative 3d but it’s really promising, you can get some cool results. The meshes are still low res and unsharp, but I hope they will keep improving it.

Nice style!

Dalle3: ‘Neon Night, painting of cyberpunk cityscape, in the style of Vincent van Gogh, luminous, energetic, moody, swirling, vibrant, neon-lit’

I think you already won! Amazing.

Hey thanks for this! I’ve been looking for some comparisons with SDXL, this is perfect. I’ve been using mostly DPM++ SDE Karras with low steps and finding it always the best, but it’s interesting that you found that at high steps it’s still high, but not a clear winner. Thanks for sharing this!

These are great! Thanks for sharing!

I had been using the bot in discord a lot for a while. The prompting is quite different to what it was in 1.5 and 2.1, you should check the Pantheon channel in discord for a good idea of the prompts that gave good results. Basically my experience is that it works very well with simple prompts, and if you want you can also get very descriptive and will get good results. Negatives are almost not needed in the bot, but I think this may be because they were injecting them already, so the 0.9 that was leaked may need them.

Hey, thanks for sharing your impressions.

I haven’t tested it yet. Are you using it with comfy UI?

From what I understand both models could be trained so the current bias may not be a big issue once people start customising them.

Have you used the bot on discord? If yes, are you finding any big differences between this and the bot? How’s it at following prompts?